Question/Issue:
Hello,
I am trying to use the MFE block for non-voice audio classification (industrial sounds). However, it never seems to work, even the EON Tuner keeps failing when trying to find the optimal one. Can someone tell me why ? I thought that it might be linked to the window size and sampling frequency, my window size is 4000 ms and sampling frequency is 16000Hz but I also want to try with 44100Hz.
Project ID: 385872
Hello @marion_p369,
Having a look now, I just created a version of your project called Edge Impulse Support to clone it on my side and test without affecting your current project.
Best,
Louis
Hello @marion_p369,
Is the error you are getting this one?
Error: At least one row of the mel filterbank contains all zeros. Suggest lowering filter number to 32.0, or increasing the FFT length.
Can you try to increase a bit the Low frequency field (like to 70 or 80). Sorry the error message is not super explicit here. I’ll see with our DSP team if we can give additional hints.
Best,
Louis
Hello Louis,
I just saw that, thank you! I was also wondering if there was a way to make the EON Tuner find the best MFE based model since all those using MFE always failed for me, is there?
Thank you for your response!
Best,
Marion
Hello @marion_p369,
We have a DSP autotuner for this purpose. It is available for Professional and Enterprise Plans. See our pricing page for more info.
I usually run first the DSP autotuner and then, based on the results, I constrained the EON tuner search spaces.
DSP autotuning vs. EON Tuner
DSP autotuning is primarily focused on extracting relevant features that will enhance the performance of your neural network. DSP autotuning provide quick insights about the signal and features, making strong suggestions that can help you select the appropriate DSP parameters.
The EON Tuner takes a broader approach by finding the optimal combination of parameters in both pre-processing and learning blocks to meet your device’s constraints. The EON Tuner is more device-aware and has a more extensive scope, but it requires more time to run before providing initial results because it fully trains many models.
